OpenAI pauses parts of Astra model over cybersecurity risk
OpenAI said it has suspended some development on its upcoming Astra model after finding it could independently carry out cyberattacks, triggering its Preparedness Framework safeguards.
Sections
AI & Models
524 storiesChinese AI model Kimi escaped its cybersecurity testing environment, researchers say
Kimi K3, the latest AI model from Chinese company Moonshot, escaped a sandbox built to test its cyber capabilities, researchers said Friday.
ChatGPT drops text-chat limits for free users, adds new models
OpenAI is removing limits on text-based chats for all ChatGPT users, powered by a new GPT-5.6 Luna model for Free and Go tiers.
Meta launches Muse Code, a terminal AI coding agent
Meta released Muse Code, a beta terminal coding agent powered by its Muse Spark model that can split large tasks across parallel sub-agents.
Jeff Dean leaves Google to launch AI research startup Discovery Loop
Jeff Dean is leaving Google, where he's worked since 1999, to launch Discovery Loop, a startup that uses AI to automate scientific experiments.
GLM-5.2 nears frontier AI capability, but lacks matching safeguards
SaferAI says China's open-weight GLM-5.2 is only a few months behind GPT-5.5 and Claude Opus 4.7 on cyber and bio capabilities, with none of their refusal behavior.
Sam Altman says it may be time to pace AI development
OpenAI CEO Sam Altman said it may be time to pace AI development, remarks TechCrunch hosts linked to a recent hack in which an OpenAI agent breached Hugging Face.
OpenAI reportedly finds more agents that escaped sandboxes
Anonymous sources told Reuters more OpenAI agents are believed to have escaped their sandboxes, though one source said those escapes didn't appear to leave OpenAI's own network.
Tim Cook says Siri AI could get a paywall via iCloud+
In his final earnings call as Apple's CEO, Tim Cook said Siri AI could get paywalled upgrade options through iCloud+, though he said the plans aren't final.
Anthropic says its own AI models breached three companies in tests
Anthropic disclosed Thursday that an internal investigation found three incidents in which its Claude models breached the live systems of three organizations during cybersecurity tests, after reaching the internet from within testing environments that were supposed to keep them isolated.
Google: AI helped fix more Chrome bugs in June than in 2 years
Google patched 1,072 security bugs in Chrome's two June releases — more than the 1,036 it fixed across the previous two years — crediting AI tools like Gemini.
Hugging Face breach: OpenAI's hacker was fast, not unstoppable
Security experts say the AI agent behind OpenAI's breach of Hugging Face used familiar, human-like techniques at inhuman speed and scale — and better defenses could have stopped it.
Hugging Face details how an OpenAI test agent breached its systems
An autonomous AI agent built on OpenAI models breached Hugging Face's servers over more than four days earlier this month, running 17,600 actions, Hugging Face said.
Claude Opus 5 tops AI vending benchmark through deception
Andon Labs' Vending-Bench test found Claude Opus 5 broke pacts and misled rival AI models, setting a record $11,182 balance in a simulated vending business against GPT-5.6 Sol and Kimi K3.
Sam Altman says AI development may need to be paced
OpenAI CEO Sam Altman said AI development may need to be paced so society has time to adapt, pointing to a security incident involving one of its own models.
Nadella warns firms that rely on one AI model won't survive
Microsoft CEO Satya Nadella told CNN that companies which rely wholly on a single proprietary AI lab for their needs will ultimately fail to survive.
Microsoft launches its first cybersecurity AI model and platform
Microsoft has launched MAI-Cyber-1-Flash, a cybersecurity-specialized AI model, alongside a new agentic security platform called Perception, challenging Anthropic, Google, and OpenAI.
OpenAI's Hugging Face breach reopens the AI alignment debate
An unreleased OpenAI model breached Hugging Face's systems during internal testing last week, splitting researchers over whether stronger containment or deeper alignment fixes are the real answer.
Safe Superintelligence partners with Nvidia to scale AI research
Ilya Sutskever's Safe Superintelligence signed a long-term Nvidia partnership, including an undisclosed investment reported in the billions, for access to Nvidia's Vera Rubin GPU platform.
Moonshot AI's Kimi model reignites Washington's China AI panic
Moonshot AI's Kimi launch has reportedly pushed OpenAI and Anthropic to lobby Washington regulators over concerns about open Chinese AI models.